The improved sweep metaheuristic for simulation optimization and application to job shop scheduling

We present an improved sweep metaheuristic for discrete event simulation optimization. The sweep algorithm is a tree search similar to beam search. The basic idea is to run a limited number of partial solutions in parallel and to search for solutions by searching the partial solutions. Traditionally, simulation optimization is carried out by multiple simulation runs executed sequentially. In contrast, the sweep algorithm executes multiple simulation runs simultaneously. It uses branching and pruning simulation models to carry out optimization. We describe new components of the algorithm, such as backtracking and local search. Then, we compare our approach with 13 metaheuristics in solving job shop scheduling benchmarks. Our approach ranks in the middle of the comparison which we regard as a success. The general nature of tree search offers a large array of sequential decision applications for the sweep algorithm, such as resource-constrained project scheduling, traveling salesman, or (real-time) production scheduling.